Fact · Speech recognition
In a 1997 review of six speech corpora, machine error rates were often more than ten times higher than human error rates on quiet, wideband, read speech.
Checked against its source on October 17, 2026
In the source's own words
“Error rates of machines are often more than an order of magnitude greater than those of humans for quiet, wideband, read speech.”
Speech recognition by machines and humans, Richard P. Lippmann, Speech Communication, vol. 22, pp. 1-15, 1997, Abstract.
Open the original- What it measures
- The review compared machine recognizers of the 1990s with human listeners on corpora that ranged from isolated digits to spontaneous telephone conversation.
- What it does not show
- Describes 1990s systems and corpora. It shows the historical gap and the shape of the problem, not the accuracy of today's engines.